0

I have negative AIC values for comparing models and want to know which would suggest a model fits better. I know that smaller values indicate a better fitting model but unsure how this works for negative models

E.g. I have one model with an AIC of -161.5795 and one with -143.8596. Does this mean the first model fits best? The R squared suggests that this is a good model with R2m = 0.55 and R2c 0.86

Olivia
  • 31
  • 1
  • 2
  • 2
    Smaller remains better, so -161.5795 is to be selected based on minimum AIC criterion. And that AIC difference is huge. – Heteroskedastic Jim Jul 01 '19 at 13:34
  • I am not sure this question is a duplicate of that question. Olivia is asking which model has more support, not if it is possible to get negative AIC values. As @Heteroskedastic Jim says, the model with the lowest value (AIC = -161.5795) should be better. However, it is always better to check the package documentation because this is not always true depending on the formulation used to compute AIC (BIC, etc.). Have a look here: https://stats.stackexchange.com/questions/195096/negative-bic-in-k-means/409739#409739 – simone Jul 01 '19 at 14:25

0 Answers0